Video Ad Captions: The Data, the Styling, and How to Add Them Accurately

A skincare brand I'll call Mara ran the same 18-second clip twice. The first cut had the founder talking to camera with clean audio and no on-screen text. The second cut was identical, except the spoken lines also appeared as large word-by-word captions in the center of the frame. Same targeting, same budget, same creative. The captioned version held viewers for noticeably longer past the three-second mark and ended up cheaper to deliver. Nothing changed except whether a muted viewer could read what the founder was saying.

That gap is the entire reason video ad captions exist as a craft rather than an afterthought. The feed does not play your sound. Captions are how you talk to the people who never tap the speaker icon, which is almost everyone.

Why video ad captions improve performance: the muted-autoplay data

Facebook, Instagram, TikTok, and LinkedIn all autoplay video silently in the feed. Sound turns on only if the viewer taps, and the data on how rarely that happens is consistent across the studies that bothered to measure it. Verizon Media and Publicis ran a 2019 study finding that 69 percent of consumers watch video with the sound off in public places and 25 percent watch muted even in private. Facebook's own internal testing, reported back in 2016 by Digiday, found that captioned video ads increased view time by an average of 12 percent. Instapage and several agency datasets since have put the share of muted Facebook video views at roughly 85 percent.

You do not need to trust a single figure. The pattern across all of them points the same direction: most feed video is consumed without audio, and the work that recovers those viewers is on-screen text. When the angle of this article promises data, this is the data. Mute is the default, captions are the recovery, and the people who do tap for sound are the minority you design for second.

This reframes what a video ad even is. It is not a 30-second spot that happens to live on a phone. It is a silent, fast-scrolling object that has to deliver its message visually first, with audio as a bonus for whoever opts in. Every line in your script that carries meaning also has to exist as text the viewer can read in a quiet room or a loud train.

There is a second, less-discussed reason captions earn their place. The World Health Organization estimates that over 1.5 billion people live with some degree of hearing loss, and far more watch in contexts where sound is socially impossible: a commute, an open-plan office, a bedroom with someone asleep nearby. Captions are how a large share of your audience receives the ad at all.

Subtitles vs captions: the distinction that matters for ads

Searchers conflate the two, so it is worth being precise. Subtitles assume the viewer can hear and translate the spoken dialogue into text, often into another language. Captions assume the viewer cannot hear and transcribe everything relevant, including the spoken words, in the original language. For feed advertising you almost always want captions, because your viewer is not hard of hearing in the medical sense, but they are functionally deaf to your ad because the platform muted it.

There is a further split that affects production. Open captions are burned into the video pixels and always visible. Closed captions are a separate track the platform can toggle on or off. For paid social, open captions win in nearly every case. You control the styling, they survive re-uploads and cross-posting, and they never depend on a viewer finding a CC button that most feed players hide. The trade-off is that burned-in text is permanent, so you cannot localize the same file later without re-rendering. For testing volume that is rarely a real constraint.

What captions do to retention

Retention is the metric that decides whether your ad gets cheap distribution. Platforms widen delivery for videos that hold attention and throttle the ones scrolled past in the first two seconds. Captions move that curve in three concrete ways.

• They make the first second legible. A scrolling viewer decides in well under a second whether to stop. Text gives them something to parse immediately: a claim, a question, a number. This is exactly why your opening caption should double as a scroll-stopping hook rather than a polite greeting. Silent video with no text asks the viewer to wait and find out, and the feed does not grant that patience.
• They keep the muted viewer in the story. Once someone stops, captions carry the narrative beat by beat. Without them, a muted viewer reaches the moment your voiceover delivers the key point, hears silence, and leaves. On a retention graph this shows up as a drop exactly where the spoken hook was supposed to land. The voiceover you produced only reaches the few who tapped; the caption reaches everyone else.
• They add a second visual layer to a still shot. Even with nothing moving on screen, animated word-by-word captions create motion that registers as activity. That micro-motion buys another beat of attention during talking-head and product-hold shots where the frame would otherwise sit static.

You can confirm all of this in your own account. Run one ad with captions and one without, then compare three-second hold rate and average watch time, two of the metrics that predict winners. The captioned cut typically holds longer because you have stopped relying on a sound channel that is switched off.

A caption styling spec that holds attention

A wall of small grey text along the bottom edge is technically captioned and practically invisible. The decisions below are the ones that change whether captions do their job. Treat the numbers as a working starting point you can adjust per brand, not a fixed rule.

• Show one to three words at a time. The word-by-word style, where words appear in sync with the voiceover and the active word is highlighted, reads faster than a static block and pulls the eye along a rhythm. That rhythm is itself a retention device.
• Size text to roughly 7 to 9 percent of the frame height. On a 1080-by-1920 vertical canvas that lands around 90 to 130 pixels of cap height for the main caption. If you squint on your own phone, it is too small. Use a heavy weight, semibold or bold, because thin weights vanish over busy footage.
• Add a stroke or plate for contrast. A 6 to 10 pixel dark outline on white text, or a semi-opaque rounded background plate at around 60 to 70 percent opacity, keeps captions legible over both bright and dark backgrounds. Pick one approach and keep it consistent.
• Respect the 9:16 safe zones. On a 1080-by-1920 canvas, the top roughly 250 pixels carry the profile and sound icons, and the bottom roughly 420 to 500 pixels carry the username, description, and CTA bar. Park your captions in the central band, around the 700 to 1300 pixel vertical range, so nothing important hides behind a Like button. The same framing rules for vertical ads apply to anything near the edges.
• Hold one typeface and one position. Captions that jump around the frame or switch fonts read as amateur and steal focus from the message. Choose a placement and keep it for the whole cut.
• Highlight the word that carries meaning. If you color or scale a word, pick the number, the benefit, or the verb. Highlighting every word highlights nothing.

What hurts retention: tiny grey text, full paragraphs that change too slowly to follow, captions that cover the speaker's mouth, and decorative animated styles so busy they compete with their own words. The goal is legibility at a glance.

How to add accurate captions to video ads

Captions only help when they are correct. A misspelled brand name or a mistimed line does more damage than no captions at all, because it signals a careless production and viewers extend that judgment to the product. The proper nouns, product names, and numbers are precisely the words that must be right and precisely the ones that get mangled. Here is a tool-agnostic method that works regardless of which editor or service you use.

1. Start from a script, not a guess. If you have the written script, paste it in as the caption source instead of letting software transcribe from scratch. Known text is correct text. If you only have the finished video, run auto-transcription first to get timing, then overwrite the words.
2. Generate a timed transcript. Use any speech-to-text engine to produce a time-coded draft. The value here is the timing, the word-by-word timestamps, not the spelling.
3. Correct the high-cost words by hand. Fix every proper noun, product name, price, and percentage against your source copy. Build a small glossary of your brand terms so you fix them the same way every time.
4. Set the chunking and timing. Break captions into one-to-three-word groups aligned to the voiceover. Each group should appear slightly before or exactly as the word is spoken, never after.
5. Apply the styling spec once, reuse it everywhere. Save your font, size, stroke, and position as a preset or template so every variant inherits the same look without redoing the work.

There are three broad ways teams put this method into practice, each with a real limitation.

• Manual captioning in an editor. Highest control, lowest throughput. Fine for one hero ad, unworkable for twenty test variants a week. Accuracy depends on the editor's attention, which fades after the fifth video.
• Auto-transcription tools. Fast, but the engine guesses at words it half-heard, and brand names, jargon, and numbers are what it gets wrong. You still proofread every line, which removes much of the speed it promised.
• Captions derived from the known script. When the system already holds the script, including cases where it wrote the script and generated the voiceover from it, the words come from text you already approved rather than a waveform. Spelling is solved. The remaining work is timing alignment, which can still drift on fast delivery or background music and is worth spot-checking. It is more tractable than re-deriving the words, but it is not zero effort.

The reason captioning matters at all for performance accounts is volume. Finding a winning creative means testing many variants, and every variant needs accurate, well-timed, correctly styled text. The method above is what keeps accuracy from becoming a per-video gamble once you scale past one ad a week.

A quick pre-launch caption review

Before any captioned ad goes live, watch it once with the sound off. If the ad stops making sense, the captions are failing and nothing else matters. Then read the first frame's text in isolation: does the opening line alone make a scrolling stranger stop, or does it say something like "Welcome to our brand" that earns no attention? Finally, recheck every proper noun and number against your source copy, confirm the captions sit inside the safe zone for each aspect ratio you export, and test legibility on the brightest and darkest footage in the cut. If the text survives both extremes and the muted version still sells, you are ready.

FAQ

Do I need captions if my ad has no voiceover?

Yes, often more so. A silent b-roll ad with no captions asks the viewer to infer your message from images alone, which rarely lands in a fast feed. Captions become the script on screen, stating the hook, the offer, and the call to action that no spoken track is carrying.

Should captions match the spoken script word for word?

For paid social, near-verbatim with light trimming is the right default. The exception is the hook: the on-screen first line can be shorter and punchier than the spoken one because text scans faster than speech. Keep the body close to the audio so sound-on viewers are not reading something different from what they hear.

What caption font and size work best for TikTok?

A heavy sans-serif at roughly 7 to 9 percent of frame height, white with a dark stroke, placed in the central band. The mechanics are similar across platforms, but TikTok and Reels reward a faster, more native word-by-word rhythm with tighter grouping, while Meta and LinkedIn tolerate slightly calmer pacing. If you are reusing one cut everywhere, the energetic word-by-word style is the safe default because it loses the least across placements.

The discipline above is correct in theory and abandoned by the third test, because captioning twenty variants by hand is genuinely tedious. Since Aitachyon derives burned-in captions from the script it already wrote, the words are never guessed, which is the one step in this whole process that does not survive repetition by hand. Try it on one ad, then watch it back with the sound off.